| name | auto-workflow |
| description | Orchestrates automated code improvement through hypothesis-driven experimentation and self-evolution |
| version | 1.1 |
| level | compound |
| molecules | ["benchmark-improver","evolution-patterns","skill-eval","sandbox-profiles"] |
metadata:
evolution-stats:
total-experiments: 870
Auto-Workflow
This skill suite orchestrates automated code improvement through systematic experimentation and self-evolution.
Skill Architecture
The auto-workflow system consists of coordinated sub-skills:
Core Pipeline
- RESEARCHER — Analyzes targets, proposes hypotheses, checks repositories
- DIRECTIVE — Strategic planning, target selection, resource allocation
- prompt-template — Structured experiment prompt construction
Quality Control
- validation-pipeline — Validates experiment outcomes against quality gates
- agent-behavior — Defines agent behavior patterns for experiment execution
Evolution
- token-efficiency — (Moved to mementum/knowledge/) Learned compression settings from experiment outcomes
Activation
Auto-workflow is triggered by the cron pipeline or manual invocation:
(auto-workflow bootstrap [target-path])
Self-Evolution
This skill auto-improves through:
- Experiment logging — Each run generates trace data
- Outcome analysis — Statistical controller learns from kept vs discarded results
- Skill refinement — Sub-skills updated based on performance data
Configuration
- Cron schedule:
0 23,3,7,11,15,19 * * *
- Pipeline lock:
var/tmp/cron/pipeline.lock
- Trace storage:
var/tmp/research-traces/
- Controller config:
var/tmp/researcher-controller.json
